Trending AI projects

**best-of-Agent-Harnesses** is a curated, ranked catalog (166 projects) of AI "agent harnesses" — the runtime layer that turns an LLM from an answerer into an actor. It includes decision guides, copy-paste templates, a ~180-line teachable harness, 10 installable "skills" that audit your own agent setup, and an MCP server/JSON/llms.txt so agents can query the list themselves. Core thesis: harness quality moves results more than model quality, and rankings barely transfer across models — so the choice is a pairing that must be re-made per model. Contributions welcome; rankings weigh relevance to harness concerns plus stars/activity.

ASPIRE is a robotics framework that replaces traditional gradient-based training with an agentic approach, where AI agents discover and refine sensorimotor skills to solve complex manipulation tasks in both simulation and the real world.

A set of AI agent skills that enable end-to-end searching, comparing, and booking of worldwide hotels and flights through natural language conversations.

Easydict is a native macOS app for fast word lookup, translation, and OCR. It auto‑detects language, offers several shortcut‑driven input modes, and aggregates results from 20+ services—including classic dictionaries, cloud translators, and LLMs like OpenAI, Gemini, Claude, and Ollama. Install via Homebrew (`brew install --cask easydict`) or a manual download. Open‑source (GPL‑3.0) and welcomes community contributions.

ArduPilot is a comprehensive, open-source autopilot software suite designed to control a wide range of autonomous vehicles, from drones and rovers to submarines.

fireworks-tech-graph is an Agent Skill for Codex and Claude Code that turns English or Chinese descriptions into polished, geometry-checked technical diagrams — SVG, high-res PNG, semantic GIF motion, and offline interactive HTML — with 12 styles and 14 diagram types.

self-llm is a Chinese-language, community-driven tutorial project that teaches beginners how to install, deploy, and fine-tune open-source large language models on Linux. It covers 50+ models (Qwen, DeepSeek, Llama, GLM, etc.), supports AMD GPUs, Huawei Ascend NPUs, and Apple Silicon, and includes four complete example projects. The recommended learning order is: environment setup → deployment → fine-tuning.

Lexora is a desktop AI agent that lets users describe a goal and then autonomously plans and executes steps across files, apps, and daily tasks. It features multi-agent collaboration, plugins, scheduled automation, and a local-first design that keeps user data and API keys on the machine.

Machine Skill is a Claude Code Agent Skill that turns a one‑line description of a real machine (e.g., “a turbofan jet engine”) into a photoreal, interactive three.js web page with accurate geometry, material rendering, and a lightweight physics solver. The page can be explored, exploded, cut‑away, or “broken”, and a deterministic 30‑second film can be recorded with a single command. Install via `npx skills add AirsupHQ/machine-skill`, then ask Claude `/machine …` to get an HTML file. The repo contains the skill manifest, a single runtime kit (no build step), physics solver recipes, example machines, and scripts for screenshots and video capture. It’s aimed at educators, creators, and engineers who need quick, physics‑backed visualisations of engines, pumps, compressors, watches, etc., but it uses simplified models (not CFD/FEA) and requires a modern GPU for the heaviest scenes.

agents‑cli is a Google‑maintained CLI (and skill bundle for AI coding assistants) that streamlines the entire lifecycle of Gemini‑based AI agents—from project scaffolding and local evaluation to cloud deployment, CI/CD, and observability.

Inbox Zero is an open‑source AI email assistant and client that organizes Gmail/Outlook, drafts replies in your style, offers bulk unsubscribe, meeting briefs, and Slack/Telegram control, plus a fast web/desktop UI. Self‑hostable via Docker and a CLI wizard.

An open-source AI coding agent for VS Code, JetBrains, and the CLI that provides access to 500+ models for code generation, planning, and debugging.

OpenCV is an open-source computer vision library that provides the tools and algorithms necessary to build AI-powered visual interpretation software.

Kami is an AI‑agent skill that converts natural‑language requests into professionally styled PDFs, PNGs, or PPTX files using eight built‑in document templates and multilingual support.

An agentic discovery engine that helps AI agents find and rank the most suitable skills for a task by analyzing real-world execution performance and user feedback instead of just popularity.

A collection of instructional skills for Claude Code that enables it to act as an automated video editing assistant within DaVinci Resolve Studio.

ShrimpReel is an open-source template for building AI-video SaaS sites that turn one photo into a viral-style video. It runs entirely on Cloudflare (Workers, D1, R2, KV, Queues) and includes user management, payments, video generation, admin tools, SEO, and compliance features. Each site is a separate repo cloned from the template, customized per keyword, and deployed independently.

A zero-code platform for auto-generating production-grade AI agents using natural language prompts, featuring multi-agent collaboration and a plug-and-play tool ecosystem.

A layout skill for AI Agents that converts Markdown into professional, style-preserved HTML specifically optimized for the WeChat Official Account editor.

A platform for continuously improving AI agents by turning production traces into versioned datasets for evaluation, optimization, and fine-tuning.

iphone-use is an automation tool that enables AI agents to control a physical iPhone by reading the screen as text and performing interactions like taps and swipes, even in apps without APIs.

A tool that converts any Git repository or local directory into a single, prompt-friendly text digest for LLMs, providing code context and token statistics.

A self-evolving ontology layer for data agents that bridges the gap between raw data and AI agents by automatically constructing and refining a semantic layer via MCP tools.

CNEquity is a local financial data lake for the Chinese market that provides structured, versioned data for research and AI agents, supporting incremental updates, quality auditing, and seamless integration with LLMs via MCP.

Casdoor is an open‑source, self‑hosted identity and access management server that provides SSO via OAuth 2.0, OIDC, SAML, CAS, LDAP, SCIM, WebAuthn, MFA, and more. It includes a web console for managing users, organizations, applications, and Casbin‑based policies, and offers SDKs, a marketplace of one‑click integrations, and deployment options via Docker, Helm, or source build.

AgentCraft turns multi-agent coding into a Minecraft world: you type a goal, a lead agent plans and workers code in isolated worktrees, while you review real diffs and approve merges from inside the game. It uses Claude Agent SDK (or Codex) and is safe because agents never push and can't touch your main checkout.

PSD2Live is an open‑source desktop suite that converts layered Photoshop files into fully rigged, animated 2‑D characters. It auto‑generates meshes, bones, physics and default motions, lets you edit everything in a canvas‑based editor, and can export to Live2D Cubism files, an open‑source Rust runtime, video, or other formats. The project also provides a local MCP service so AI agents can programmatically read or modify a project.

Symphony is a framework for turning project work into autonomous implementation runs, allowing teams to manage high-level work instead of supervising coding agents.

A curated collection of 300+ Antigravity “skills” (workflow packages) ported from Claude Code Agents, with an npm CLI for searching, installing, and managing them in Google Antigravity’s AI‑agent environment.

LMCache is a vendor-neutral KV cache management layer that offloads and persists LLM inference cache to improve throughput and reduce latency for long-context and agentic workloads.

json-render is a generative UI framework that lets AI create interfaces from prompts while staying safe and predictable. It works by constraining AI output to a catalog of defined components and actions, which are then rendered across many platforms.

Noise is a competitive speedrun for pretraining a fixed transformer without backpropagation. It provides track rules, a fixed dataset and model, and a leaderboard to benchmark learning algorithms that only use forward passes, aiming to beat backprop-based training.

A method-native AI work system that provides a CLI workbench and SDK to make complex software development tasks reliable, recoverable, and verifiable.

A collection of standardized, portable skills for AI coding agents, covering tasks from frontend review to cloud infrastructure, supported by an evaluation CLI for testing and iteration.

A native desktop installer and manager for AI agent personas that synchronizes specialist personas across various AI coding tools like Cursor and Claude Code.

DeepWiki-Open is an automated tool that analyzes code repositories to generate interactive wikis, visual diagrams, and guided codemaps for documentation.

A facial recognition tool for macOS that allows users to unlock their Mac using their face by automatically typing the stored password.

A project context and engineering control plane that provides AI coding agents with durable project knowledge, governance, and risk-driven quality gates.

An MCP server that gives AI coding agents access to Figma design data, allowing them to implement designs more accurately than using screenshots.

TryPost is an open‑source, AGPL‑licensed social‑media scheduler that offers both a web UI for human editors and an MCP/REST API for AI agents to create, queue, publish, and analyze posts across dozens of networks.

stable‑diffusion.cpp is a pure C/C++ inference engine for diffusion‑based image (and some video) models. Built on ggml, it runs on CPU, CUDA, Vulkan, Metal, OpenCL, and SYCL, supports dozens of models (Stable Diffusion, FLUX, Qwen‑Image, etc.), multiple weight formats, and extensions like LoRA, ControlNet, and upscaling. It offers pre‑built binaries, a one‑line CLI, and a rich ecosystem of language bindings and UI front‑ends.

ReachAI is an open‑source Java‑centric platform that retrofits existing enterprise apps (OA, ERP, CRM, HR, etc.) with AI assistants. It scans UI and APIs, registers business methods as annotated Capabilities, uses LLM‑driven “AI Coding” to implement suggested refactors, and deploys Agents and versioned Workflows that can answer natural‑language queries, fill forms and trigger transactions—all while preserving original authentication, requiring user confirmation for writes, and providing full audit traces.

An in-memory, GPU-accelerated SQL relational database and visualization platform designed for high-speed analysis of massive geospatial and relational datasets.

A comprehensive, version‑controlled guide (with website, Markdown source, and optional MCP server) for using Anthropic’s Claude Code coding assistant—covering installation, agent harness design, security, observability, and enterprise adoption.

An autonomous AI development workspace for Android that lets users chat with coding agents, execute Linux commands, and build apps directly on their phone.

An agentic video creation framework that automates the entire pipeline from narrative planning and scriptwriting to storyboarding and final video generation.

A modern open-source Linux server management panel that provides a web GUI for server operations and a unified platform for managing AI models and agents.

A distributed search and analytics engine and vector database that enables near real-time search and analysis over massive datasets for AI and monitoring applications.